Project Summary Decades of psychology research have shown that working memory is limited, and humans can only hold a few items in mind at the same time. However, cognitive tasks like planning and problem solving require access to many pieces of information at once. To overcome this constraint, we enlist mnemonic strategies, for instance grouping pieces of information into chunks, as we commonly do to remember telephone or social security numbers. Mnemonic chunking allows us to flexibly organize information on line, providing a fundamental building block for advanced cognitive abilities. Chunking impairments occur when damage or dysfunction involves the dorsolateral prefrontal cortex (dlPFC), for instance in patients with schizophrenia, and severely compromises overall cognitive function. Thus, determining how the brain organizes information is a necessary step toward understanding the mechanisms of advanced cognition, and how these go awry in disease states. A key challenge is that strategies for organizing information are self-generated and highly variable in a laboratory setting. A central innovation of this proposal is the novel computational approach used to identify spontaneous mnemonic chunking in macaque monkeys. This is critical because animal models allow us to interrogate brain function with advanced neurophysiological tools. Here, we will use high-density, multi-site recording and targeted neuromodulation to understand the circuit mechanisms that chunk mnemonic information. Previous theoretical work suggests that chunks arise from compressed working memory representations that act as neural shorthand, economizing on processing resources at the cost of degrading some original information. Neurons in dlPFC encode items in working memory, and their dynamics are shaped by recurrent interactions with the basal ganglia. Thus, we hypothesize that corticostriatal interactions promote the efficient reorganization of working memory that underlies chunking. To test this we will investigate dlPFC- striatal dynamics when monkeys spontaneously chunk information in a self-organized working memory task. We will record large numbers of single neurons and local field potentials, and dynamically decode representations held in working memory to assess how mnemonic codes and corticostriatal interactions change when items are or are not chunked. In addition, exogenous stimulation will test the causal role of striatal circuits in promoting the formation of mnemonic chunks. Together, these experiments will determine how the brain establishes mnemonic chunks to optimize working memory performance. This will shed light on a fundamental feature of advanced cognition, and how dysfunction in these mechanisms could give rise to disorders of thought and memory. Finally, understanding mechanisms that optimize cognitive function in a biological system may fuel creative advances that optimize performance in artificial intelligence systems.